Junie CLI: "The task was stopped due to an error" for all OpenRouter BYOK models

### Summary
Every model I select through Junie CLI's built-in OpenRouter provider fails immediately when running any prompot with "The task was stopped due to an error. Try again." This happens across multiple, unrelated models on OpenRouter, so it doesn't appear to be model-specific. Native JetBrains AI-provided models work fine; the issue is isolated to the OpenRouter BYOK path.

### Environment
OS: CachyOS Linux
Junie CLI version: Junie version: 26.7.27 (2470.4)
Launched via: Normal Terminal and JetBrains Rider integrated terminal
Rider version: 2026.2.0.2
Connection method: /account -> "Manage your API keys" -> OpenRouter -> [pasted API key]

### Steps to Reproduce
Run Junie CLI from the Rider terminal on Linux.
Run /account, select "Manage your API keys" select OpenRouter, paste a valid OpenRouter API key.
Run /model, select an OpenRouter-provided model (tested: Grok 4.5, GPT-5.6-Luna, GPT-5.6-Terra).
Send any prompt

### Expected Behavior
The task runs normally against the selected OpenRouter model, same as it does for JetBrains AI-provided models.

### Actual Behavior
The task immediately stops with: "The task was stopped due to an error. Try again."

### Models Tested (all fail identically)
Grok 4.5 (OpenRouter)
GPT-5.6-Luna (OpenRouter)
GPT-5.6-Terra (OpenRouter)

### Additional Notes
The OpenRouter API key is confirmed valid and working; tested independently outside Junie.
Native JetBrains AI-subscription models work without issue in the same Junie CLI install, so this is isolated to the OpenRouter BYOK integration specifically.
Since the failure reproduces across two unrelated model providers routed through OpenRouter, this looks like a general issue in how Junie CLI handles OpenRouter's response, rather than a per-model problem.

Start by reproducing the failure through /account, the OpenRouter API-key flow, and /model with a listed OpenRouter model on Linux. Trace the OpenRouter request and response handling, then verify that prompts complete successfully while native JetBrains AI models remain unaffected.

Written by the indexing model from the issue text.